Every year between 5 and 20 percent of the US population contracts influenza. In a typical year, complications from the flu lead to 200,000 hospitalizations and 30,000 deaths. The influenza vaccine has the potential to greatly reduce this public health burden. Randomized controlled trials demonstrate that vaccination substantially reduces the probability that a person develops influenza, but they lack sufficient power to estimate the vaccine's efficacy in reducing rare events such as hospitalizations and deaths. Comparisons between the vaccinated and unvaccinated using observational data suggest that the vaccine reduces hospitalizations and deaths. However, these estimates suffer from selection bias. We propose two quasi- experimental approaches to estimating the influenza vaccine's effect on flu-related doctors' visits, hospitalizations, and deaths. The first approach exploits age-based guidelines for influenza vaccination. The guidelines generate a large and abrupt increase in vaccination rates when people reach the age threshold. This sharp increase allows us to implement a regression discontinuity design to estimate the effectiveness of the vaccine. The core insight in this approach is that, despite the increase in vaccination rates, individuals just under the age threshold are a plausible comparison group for those just over the age threshold. The second approach leverages quasi-random year-to-year variation in how well the vaccine matches circulating influenza strains to generate variation in effective vaccination rates. The essential insight in this approach is that people who get vaccinated in years with a poor match provide a plausible comparison group for people who get vaccinated in years with a good match. We have the following three specific aims. First, estimate the influenza vaccine's effectiveness in protecting individuals near age 65 from influenza severe enough to result in a visit to the doctor, hospitalization, or death. Second, document the influenza vaccine's effect on reducing influenza-related morbidity and mortality among individuals at broader age ranges. Third, determine if current influenza vaccination rates significantly reduce morbidity and mortality among the unvaccinated through herd effects. Estimates of herd effects can be used in conjunction with estimates of the direct effects of vaccination from the first two aims to compare current vaccination guidelines to proposed alternatives. Depending on the relative sizes of the vaccine's direct effect and the herd effect, it may be preferable to target groups that are at low risk of developing severe illness but are at high risk of exposing others rather than the current policy which focuses on groups at high risk of developing severe illness.